Federated Learning Modem Algorithm Training Edge Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern modem algorithms in edge devices are difficult to train continuously due to security and privacy concerns, as well as high bandwidth and power consumption, when sending data collected from actual use for training purposes.
Innovation Solution
Implementing federated learning by using a local controller within the edge device to receive inputs, update parameters, and share only the updated weights or gradients with a global server, allowing for secure and efficient training of modem algorithms without exposing sensitive information or increasing bandwidth and power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data collected from actual use is sent over the network for training, then the modem algorithm can be improved, but user privacy is violated and bandwidth is consumed
Solution Approach 1:
The patent extracts only the essential training information (gradients or weight updates) from the edge device while leaving the raw collected data localized and private. This extraction approach allows algorithm improvement without transmitting sensitive user information over the network.
Solution Approach 2:
The patent introduces an intermediary training process where local controllers train the modem algorithm using collected data locally, then share only the trained model parameters or gradients with the global server. This intermediary step prevents direct exposure of raw user data while still enabling algorithm improvement.
2Productivity
If access to the modem algorithm is allowed for training purposes, then the algorithm can be improved, but proprietary information is revealed
Solution Approach 1:
The patent extracts only the necessary updates (gradients or weight changes) from the local controller's training process and transmits these to the global server. The complete proprietary algorithm remains localized and secure, while only incremental improvement data is shared.
Solution Approach 2:
The patent segments the training process into local and global components. The local controller performs training locally with proprietary data, then shares only the segmented results (updates) with the global server, maintaining security while enabling collaborative improvement.
3Productivity
If data is sent for centralized training, then the algorithm can be improved, but bandwidth consumption increases
Solution Approach 1:
The patent extracts only the essential training signals (gradients or weight updates) which are significantly smaller in size than the raw collected data. This extraction dramatically reduces the amount of data that needs to be transmitted over the network.
Solution Approach 2:
Instead of sending raw data from edge devices to the server for training, the patent inverts the approach by having the server receive only the trained model updates. This reversal of the data flow direction significantly reduces communication overhead.
4Productivity
If data is sent for centralized training, then the algorithm can be improved, but power consumption increases
Solution Approach 1:
The patent extracts only the necessary training information (gradients or weight updates) rather than transmitting complete datasets. This extraction significantly reduces the computational and communication energy required for training.
Solution Approach 2:
The patent inverts the traditional centralized training model by performing the computationally intensive training process locally at the edge device and only communicating the results. This inversion reduces the energy consumption associated with data transmission and centralized processing.
Data Source
AI summary
A system and a method are disclosed, the method including receiving, by a first local controller of a first edge device, an input associated with an environment in which the first edge device operates, using a first machine-learning algorithm, determining, by the first local controller, a parameter for a pre-trained modem algorithm of the first edge device based on the input, executing a task on the first edge device based on executing the pre-trained modem algorithm with the parameter, determining a result of executing the task, training the first machine-learning algorithm, generating a first update to the first machine-learning algorithm based on the training, sending the first update to a server, receiving, from the server, a server update to the first machine-learning algorithm, and based on the server update, updating the first machine-learning algorithm.